Aim of this study was to compare three different working definitions for selecting a frail elderly population. Frailty was defined as inactivity combined with (1) low energy intake (n = 29), (2) weight loss (n = 26), or (3) low body mass index (n = 26). In the Zutphen Elderly Study (n = 450; age 69-89 years) differences in health, functioning, and diet in 1990 and functional decline and mortality in the following 3 years between "frail" and "nonfrail" participants, according to the working definitions, were studied using logistic regression analysis. Differences according to the inactivity/weight loss criterium were more pronounced than according to the other two criteria. Inactivity/weight loss was associated with lower subjective health and performance and more diseases and disabilities in 1990. Three-year relative risks of mortality (odds ratio [OR]: 4.1, 1.8-9.4) and functional decline (OR: 5.2, 1.04-25.8 for disabilities, OR: 3.7, 0.8-16.2 for performance) were higher as well. Inactivity in combination with weight loss seems a practicable working definition for selecting a frail elderly population.